feat: legacy features to ruby llm (#12994)

This commit is contained in:
Aakash Bakhle
2025-12-11 14:17:28 +05:30
committed by GitHub
parent f2054e703a
commit 1de8d3e56d
39 changed files with 860 additions and 755 deletions

View File

@@ -7,14 +7,6 @@ module Integrations::LlmInstrumentation
include Integrations::LlmInstrumentationHelpers
include Integrations::LlmInstrumentationSpans
PROVIDER_PREFIXES = {
'openai' => %w[gpt- o1 o3 o4 text-embedding- whisper- tts-],
'anthropic' => %w[claude-],
'google' => %w[gemini-],
'mistral' => %w[mistral- codestral-],
'deepseek' => %w[deepseek-]
}.freeze
def instrument_llm_call(params)
return yield unless ChatwootApp.otel_enabled?
@@ -66,16 +58,57 @@ module Integrations::LlmInstrumentation
end
end
def determine_provider(model_name)
return 'openai' if model_name.blank?
def instrument_embedding_call(params)
return yield unless ChatwootApp.otel_enabled?
model = model_name.to_s.downcase
PROVIDER_PREFIXES.each do |provider, prefixes|
return provider if prefixes.any? { |prefix| model.start_with?(prefix) }
instrument_with_span(params[:span_name] || 'llm.embedding', params) do |span, track_result|
set_embedding_span_attributes(span, params)
result = yield
track_result.call(result)
set_embedding_result_attributes(span, result)
result
end
end
'openai'
def instrument_audio_transcription(params)
return yield unless ChatwootApp.otel_enabled?
instrument_with_span(params[:span_name] || 'llm.audio.transcription', params) do |span, track_result|
set_audio_transcription_span_attributes(span, params)
result = yield
track_result.call(result)
set_transcription_result_attributes(span, result)
result
end
end
def instrument_moderation_call(params)
return yield unless ChatwootApp.otel_enabled?
instrument_with_span(params[:span_name] || 'llm.moderation', params) do |span, track_result|
set_moderation_span_attributes(span, params)
result = yield
track_result.call(result)
set_moderation_result_attributes(span, result)
result
end
end
def instrument_with_span(span_name, params, &)
result = nil
executed = false
tracer.in_span(span_name) do |span|
track_result = lambda do |r|
executed = true
result = r
end
yield(span, track_result)
end
rescue StandardError => e
ChatwootExceptionTracker.new(e, account: resolve_account(params)).capture_exception
raise unless executed
result
end
private
@@ -86,36 +119,4 @@ module Integrations::LlmInstrumentation
nil
end
def setup_span_attributes(span, params)
set_request_attributes(span, params)
set_prompt_messages(span, params[:messages])
set_metadata_attributes(span, params)
end
def record_completion(span, result)
if result.respond_to?(:content)
span.set_attribute(ATTR_GEN_AI_COMPLETION_ROLE, result.role.to_s) if result.respond_to?(:role)
span.set_attribute(ATTR_GEN_AI_COMPLETION_CONTENT, result.content.to_s)
elsif result.is_a?(Hash)
set_completion_attributes(span, result) if result.is_a?(Hash)
end
end
def set_request_attributes(span, params)
provider = determine_provider(params[:model])
span.set_attribute(ATTR_GEN_AI_PROVIDER, provider)
span.set_attribute(ATTR_GEN_AI_REQUEST_MODEL, params[:model])
span.set_attribute(ATTR_GEN_AI_REQUEST_TEMPERATURE, params[:temperature]) if params[:temperature]
end
def set_prompt_messages(span, messages)
messages.each_with_index do |msg, idx|
role = msg[:role] || msg['role']
content = msg[:content] || msg['content']
span.set_attribute(format(ATTR_GEN_AI_PROMPT_ROLE, idx), role)
span.set_attribute(format(ATTR_GEN_AI_PROMPT_CONTENT, idx), content.to_s)
end
end
end

View File

@@ -0,0 +1,88 @@
# frozen_string_literal: true
module Integrations::LlmInstrumentationCompletionHelpers
include Integrations::LlmInstrumentationConstants
private
def set_embedding_span_attributes(span, params)
span.set_attribute(ATTR_GEN_AI_PROVIDER, determine_provider(params[:model]))
span.set_attribute(ATTR_GEN_AI_REQUEST_MODEL, params[:model])
span.set_attribute('embedding.input_length', params[:input]&.length || 0)
span.set_attribute(ATTR_LANGFUSE_OBSERVATION_INPUT, params[:input].to_s)
set_common_span_metadata(span, params)
end
def set_audio_transcription_span_attributes(span, params)
span.set_attribute(ATTR_GEN_AI_PROVIDER, 'openai')
span.set_attribute(ATTR_GEN_AI_REQUEST_MODEL, params[:model] || 'whisper-1')
span.set_attribute('audio.duration_seconds', params[:duration]) if params[:duration]
span.set_attribute(ATTR_LANGFUSE_OBSERVATION_INPUT, params[:file_path].to_s) if params[:file_path]
set_common_span_metadata(span, params)
end
def set_moderation_span_attributes(span, params)
span.set_attribute(ATTR_GEN_AI_PROVIDER, 'openai')
span.set_attribute(ATTR_GEN_AI_REQUEST_MODEL, params[:model] || 'text-moderation-latest')
span.set_attribute('moderation.input_length', params[:input]&.length || 0)
span.set_attribute(ATTR_LANGFUSE_OBSERVATION_INPUT, params[:input].to_s)
set_common_span_metadata(span, params)
end
def set_common_span_metadata(span, params)
span.set_attribute(ATTR_LANGFUSE_USER_ID, params[:account_id].to_s) if params[:account_id]
span.set_attribute(ATTR_LANGFUSE_TAGS, [params[:feature_name]].to_json) if params[:feature_name]
end
def set_embedding_result_attributes(span, result)
span.set_attribute('embedding.dimensions', result&.length || 0) if result.is_a?(Array)
span.set_attribute(ATTR_LANGFUSE_OBSERVATION_OUTPUT, "[#{result&.length || 0} dimensions]")
end
def set_transcription_result_attributes(span, result)
transcribed_text = result.respond_to?(:text) ? result.text : result.to_s
span.set_attribute('transcription.length', transcribed_text&.length || 0)
span.set_attribute(ATTR_LANGFUSE_OBSERVATION_OUTPUT, transcribed_text.to_s)
end
def set_moderation_result_attributes(span, result)
span.set_attribute('moderation.flagged', result.flagged?) if result.respond_to?(:flagged?)
span.set_attribute('moderation.categories', result.flagged_categories.to_json) if result.respond_to?(:flagged_categories)
output = {
flagged: result.respond_to?(:flagged?) ? result.flagged? : nil,
categories: result.respond_to?(:flagged_categories) ? result.flagged_categories : []
}
span.set_attribute(ATTR_LANGFUSE_OBSERVATION_OUTPUT, output.to_json)
end
def set_completion_attributes(span, result)
set_completion_message(span, result)
set_usage_metrics(span, result)
set_error_attributes(span, result)
end
def set_completion_message(span, result)
message = result[:message] || result.dig('choices', 0, 'message', 'content')
return if message.blank?
span.set_attribute(ATTR_GEN_AI_COMPLETION_ROLE, 'assistant')
span.set_attribute(ATTR_GEN_AI_COMPLETION_CONTENT, message)
end
def set_usage_metrics(span, result)
usage = result[:usage] || result['usage']
return if usage.blank?
span.set_attribute(ATTR_GEN_AI_USAGE_INPUT_TOKENS, usage['prompt_tokens']) if usage['prompt_tokens']
span.set_attribute(ATTR_GEN_AI_USAGE_OUTPUT_TOKENS, usage['completion_tokens']) if usage['completion_tokens']
span.set_attribute(ATTR_GEN_AI_USAGE_TOTAL_TOKENS, usage['total_tokens']) if usage['total_tokens']
end
def set_error_attributes(span, result)
error = result[:error] || result['error']
return if error.blank?
span.set_attribute(ATTR_GEN_AI_RESPONSE_ERROR, error.to_json)
span.status = OpenTelemetry::Trace::Status.error(error.to_s.truncate(1000))
end
end

View File

@@ -2,38 +2,52 @@
module Integrations::LlmInstrumentationHelpers
include Integrations::LlmInstrumentationConstants
include Integrations::LlmInstrumentationCompletionHelpers
def determine_provider(model_name)
return 'openai' if model_name.blank?
model = model_name.to_s.downcase
LlmConstants::PROVIDER_PREFIXES.each do |provider, prefixes|
return provider if prefixes.any? { |prefix| model.start_with?(prefix) }
end
'openai'
end
private
def set_completion_attributes(span, result)
set_completion_message(span, result)
set_usage_metrics(span, result)
set_error_attributes(span, result)
def setup_span_attributes(span, params)
set_request_attributes(span, params)
set_prompt_messages(span, params[:messages])
set_metadata_attributes(span, params)
end
def set_completion_message(span, result)
message = result[:message] || result.dig('choices', 0, 'message', 'content')
return if message.blank?
span.set_attribute(ATTR_GEN_AI_COMPLETION_ROLE, 'assistant')
span.set_attribute(ATTR_GEN_AI_COMPLETION_CONTENT, message)
def record_completion(span, result)
if result.respond_to?(:content)
span.set_attribute(ATTR_GEN_AI_COMPLETION_ROLE, result.role.to_s) if result.respond_to?(:role)
span.set_attribute(ATTR_GEN_AI_COMPLETION_CONTENT, result.content.to_s)
elsif result.is_a?(Hash)
set_completion_attributes(span, result)
end
end
def set_usage_metrics(span, result)
usage = result[:usage] || result['usage']
return if usage.blank?
span.set_attribute(ATTR_GEN_AI_USAGE_INPUT_TOKENS, usage['prompt_tokens']) if usage['prompt_tokens']
span.set_attribute(ATTR_GEN_AI_USAGE_OUTPUT_TOKENS, usage['completion_tokens']) if usage['completion_tokens']
span.set_attribute(ATTR_GEN_AI_USAGE_TOTAL_TOKENS, usage['total_tokens']) if usage['total_tokens']
def set_request_attributes(span, params)
provider = determine_provider(params[:model])
span.set_attribute(ATTR_GEN_AI_PROVIDER, provider)
span.set_attribute(ATTR_GEN_AI_REQUEST_MODEL, params[:model])
span.set_attribute(ATTR_GEN_AI_REQUEST_TEMPERATURE, params[:temperature]) if params[:temperature]
end
def set_error_attributes(span, result)
error = result[:error] || result['error']
return if error.blank?
def set_prompt_messages(span, messages)
messages.each_with_index do |msg, idx|
role = msg[:role] || msg['role']
content = msg[:content] || msg['content']
span.set_attribute(ATTR_GEN_AI_RESPONSE_ERROR, error.to_json)
span.status = OpenTelemetry::Trace::Status.error(error.to_s.truncate(1000))
span.set_attribute(format(ATTR_GEN_AI_PROMPT_ROLE, idx), role)
span.set_attribute(format(ATTR_GEN_AI_PROMPT_CONTENT, idx), content.to_s)
end
end
def set_metadata_attributes(span, params)

View File

@@ -1,7 +1,6 @@
# frozen_string_literal: true
require 'opentelemetry_config'
require_relative 'llm_instrumentation_constants'
module Integrations::LlmInstrumentationSpans
include Integrations::LlmInstrumentationConstants

View File

@@ -77,21 +77,22 @@ class Integrations::Openai::ProcessorService < Integrations::LlmBaseService
end
def add_message_if_within_limit(character_count, message, messages, in_array_format)
if valid_message?(message, character_count)
add_message_to_list(message, messages, in_array_format)
character_count += message.content.length
content = message.content_for_llm
if valid_message?(content, character_count)
add_message_to_list(message, messages, in_array_format, content)
character_count += content.length
[character_count, true]
else
[character_count, false]
end
end
def valid_message?(message, character_count)
message.content.present? && character_count + message.content.length <= TOKEN_LIMIT
def valid_message?(content, character_count)
content.present? && character_count + content.length <= TOKEN_LIMIT
end
def add_message_to_list(message, messages, in_array_format)
formatted_message = format_message(message, in_array_format)
def add_message_to_list(message, messages, in_array_format, content)
formatted_message = format_message(message, in_array_format, content)
messages.prepend(formatted_message)
end
@@ -99,17 +100,17 @@ class Integrations::Openai::ProcessorService < Integrations::LlmBaseService
in_array_format ? [] : ''
end
def format_message(message, in_array_format)
in_array_format ? format_message_in_array(message) : format_message_in_string(message)
def format_message(message, in_array_format, content)
in_array_format ? format_message_in_array(message, content) : format_message_in_string(message, content)
end
def format_message_in_array(message)
{ role: (message.incoming? ? 'user' : 'assistant'), content: message.content }
def format_message_in_array(message, content)
{ role: (message.incoming? ? 'user' : 'assistant'), content: content }
end
def format_message_in_string(message)
def format_message_in_string(message, content)
sender_type = message.incoming? ? 'Customer' : 'Agent'
"#{sender_type} #{message.sender&.name} : #{message.content}\n"
"#{sender_type} #{message.sender&.name} : #{content}\n"
end
def summarize_body

17
lib/llm_constants.rb Normal file
View File

@@ -0,0 +1,17 @@
# frozen_string_literal: true
module LlmConstants
DEFAULT_MODEL = 'gpt-4.1-mini'
DEFAULT_EMBEDDING_MODEL = 'text-embedding-3-small'
PDF_PROCESSING_MODEL = 'gpt-4.1-mini'
OPENAI_API_ENDPOINT = 'https://api.openai.com'
PROVIDER_PREFIXES = {
'openai' => %w[gpt- o1 o3 o4 text-embedding- whisper- tts-],
'anthropic' => %w[claude-],
'google' => %w[gemini-],
'mistral' => %w[mistral- codestral-],
'deepseek' => %w[deepseek-]
}.freeze
end

View File

@@ -1,8 +0,0 @@
# frozen_string_literal: true
module OpenAiConstants
DEFAULT_MODEL = 'gpt-4.1-mini'
DEFAULT_ENDPOINT = 'https://api.openai.com'
DEFAULT_EMBEDDING_MODEL = 'text-embedding-3-small'
PDF_PROCESSING_MODEL = 'gpt-4.1-mini'
end